Unsupervised Neural Network for NMR Relaxation Inversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional NMR relaxation time spectrum inversion methods are prone to uncertainty due to noise in the original spin relaxation signal, requiring manual adjustment of regularization parameters and being sensitive to prior information, while supervised deep learning methods rely on labeled data, making them impractical for experimental data.
Innovation Solution
An unsupervised neural network is trained using unlabeled NMR relaxation signals with a defined loss function that learns regularization parameters autonomously, allowing for the inversion of NMR relaxation time spectra without reference to labeled data, thereby reducing dependence on manual parameter adjustments and enabling real-time online monitoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Laplace inversion method is used to obtain NMR relaxation time spectrum, then the inversion can be performed, but the inversion results are uncertain due to noise and sensitivity to regularization parameters
Solution Approach 1:
The neural network performs self-learning through unsupervised training, automatically determining regularization parameters without manual intervention. The network learns optimal parameters by processing training data and minimizing reconstruction error, enabling it to serve itself in parameter optimization rather than requiring external manual tuning.
Solution Approach 2:
The invention transforms the inversion problem from a traditional mathematical approach to a machine learning approach by changing the parameters from fixed regularization values to learnable neural network weights. The network dynamically adjusts effective regularization parameters based on the input signal characteristics, improving both accuracy and reliability.
2Ease of operation
If supervised deep learning method is used for inversion, then manual parameter adjustment is reduced, but labeled training data is required which is impractical for experimental data
Solution Approach 1:
Instead of using the conventional supervised approach where labels are required, the invention inverts the training paradigm by using unsupervised learning. The network learns from unlabeled experimental data alone, eliminating the need for difficult-to-obtain labeled training data while maintaining ease of operation.
Solution Approach 2:
The neural network performs self-learning through unsupervised training, automatically determining regularization parameters without manual intervention. The network learns optimal parameters by processing training data and minimizing reconstruction error, enabling it to serve itself in parameter optimization rather than requiring external manual tuning.
3Measurement precision
If nonlinear inversion methods with multiple iterations are used, then global minimum can be found, but the operation speed is very slow
Solution Approach 1:
The neural network is pre-trained on a large dataset during an offline training phase, learning the underlying patterns and relationships. When actual inversion is needed, the pre-trained network provides rapid predictions without requiring multiple iterative calculations, thus achieving both accuracy and speed.
Solution Approach 2:
The invention creates a computational model (neural network) that copies and learns from numerous example inversion problems during training. Once trained, this copied knowledge enables rapid solving of new inversion problems without repeating the complex iterative optimization process for each case.
Data Source
AI summary
An NMR relaxation time inversion method based on an unsupervised neural network includes simulating inversion kernel matrix, simulating continuous NMR relaxation time spectrum, simulating noise, calculating NMR relaxation signals as samples, various samples forming a sample set, constructing an unsupervised neural network model, and defining a loss function of the unsupervised neural network model; and taking the samples in the training sample set as an input of the unsupervised neural network model, to obtain an optimal mapping relationship between the NMR relaxation signals and the NMR relaxation time spectrum with a minimum loss function. The present invention provides the possibility of using experimental data as the sample for training since the trading sample does not need to be labeled, can automatically learn the optimal regularization parameters without depending on the initial value and manual experience, and predicts fast.